Weighted recursive denoising deep neural network and method for multilayer convolution sparse coding

Abstract

The invention provides an end-to-end weighted recursive denoising convolutional neural network WRDnCN-LISTA + based on a multi-layer convolutional sparse coding (MLCSC) model, and the model is based on an expanded multi-layer learning iterative soft threshold algorithm (ML-LISTA), and introduces a recursive weighted supervision mechanism to realize natural image denoising. The expanded ML-LISTA algorithm can be in one-to-one correspondence with the convolutional neural network, and the weighted supervision mechanism can also improve the degradation problem brought by a simple recursive structure in the deep network. The introduced learnable weight uses output results of all intermediate recursions, so that the influence of different recursion times on the network performance is weakened, and the denoising performance is also enhanced. The specific parameter sharing property of the recursive network enables the parameter cost consumed for constructing a deep convolutional neural network to be reduced, and ensures that all parameters in the model can be adaptively updated by using back propagation through minimizing a loss function.

Description

technical field [0001] The invention relates to the technical field of image processing, more specifically, it relates to a weighted recursive denoising deep neural network and method of multi-layer convolution sparse coding. Background technique [0002] Image denoising is a crucial image processing problem in the field of computer vision. It is regarded as a preprocessing step for advanced vision problems and is widely used in practical applications such as medical image analysis, remote sensing imaging, and digital photography. Among them, the removal of additive Gaussian white noise (AGWN) from synthetic images has attracted a lot of research. This denoising task assumes that observed images contaminated by noise Y=X+V, where X is a potential clean image and V represents a known standard Additive white Gaussian noise with variance σ, through a denoising algorithm to obtain a potentially clean image from the polluted observation image Y. [0003] Existing image denoising...

AI Technical Summary

Problems solved by technology

However, the training of the recursive network is difficult. As the number of recursions increases, the depth of the network will gradually deepen. When the image feature information acquired at the shallow layer is transmitted to the deep layer of the network through a single recursive structure, it is easy to cause the problem of incomplete information utilization.

At the same time, learning between distant pixels will become difficult, and it may also face potential gradient disappearance and gradient explosion problems

Images